Instance Segmentation Model Weights

Light weight deep learning models for instance image segmentation.

Overview

Mask RCNN is a state-of-art deep learning model for instance image segmentation, where the goal is to assign instance level labels ( e.g. person1, person2, cat) to every pixel in an input image. Mask RCNN algorithm is available in the TensorFlow Model Garden which is a repository with a number of different implementations of state-of-the-art models and modeling solutions for TensorFlow users.

Model Categories

Material model - Detects the high level category (e.g. plastic, paper, etc) of an object according to its material type.
Material Form model - Detects the category of the of an object according to its physical product formation (e.g. cup, plate, pen, etc).
Plastic model - Detects the category of a object according to its plastic types (e.g. HDPE, LDPE, etc)

The goal to develop these models is to bring transparency & traceability in the world of waste recycling.

Model paths in GCP buckets

Model categories	Model backbone	Model type	GCP bucket path
Material Model	Resnet	saved model	click here
		TFLite	click here
Material Form model	Resnet	saved model	click here
		TFLite	click here
Plastic model	Resnet	saved model	click here
		TFLite	click here

Installation & Download

You need to use gsutil command line tool to download the deep learning weights from their respective GCP buckets.

Start by logging into the Google Cloud Console and create a project.
Download and install the Google Cloud SDK from the official website.
You can verity that the install went successfully by opening up a machine terminal and executing the command

gsutil -v

Once your install is successfull you can download the weight from the GCP buckets using gsutil command. For example -

gsutil -m cp -r gs://official_release/version_1/plastic_types_model/saved_model/* .

The above command will download the weights in your local directory.